The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies (19 page)

If we look more closely at the jobs eliminated as companies reorganized,
skill-biased technical change
can be a bit of a misleading moniker. In particular, it would be a mistake to assume that all ‘college-level tasks’ are hard to automate while ‘kindergarten tasks’ are easy. In recent years, ‘low-skill tasks’ haven’t always been the ones being automated; more often it has been ‘tasks that machines can do better than humans.’ Of course, that’s a bit of a tautology, but a useful tautology nonetheless. Repetitive work on an assembly line is easier to automate than the work of a janitor. Routine clerical work like processing payments is easier to automate than handling customers’ questions. At present, machines are not very good at walking up stairs, picking up a paperclip from the floor, or reading the emotional cues of a frustrated customer.

To capture these distinctions, work by our MIT colleagues Daron Acemoglu and David Autor suggests that work can be divided into a two-by-two matrix: cognitive versus manual and routine versus nonroutine.
²⁵
They found that the demand for work has been falling most dramatically for routine tasks, regardless of whether they are cognitive or manual. This leads to job polarization: a collapse in demand for middle-income jobs, while nonroutine cognitive jobs (such as financial analysis) and nonroutine manual jobs (like hairdressing) have held up relatively well.

Building on Acemoglu and Autor’s work, economists Nir Jaimovich of Duke University and Henry Siu of the University of British Columbia found a link between job polarization and the jobless recoveries that have defined the last three recessions. For most of the nineteenth and twentieth centuries, employment usually rebounded strongly after each recession, but since the 1990s employment didn’t recover briskly after recessions. Again, it’s not a coincidence that as the computerization of the economy advanced, post-recession hiring patterns changed. When Jaimovich and Siu compared the 1980s, 1990s, and 2000s, they found that the demand for routine cognitive tasks such as cashiers, mail clerks, and bank tellers and routine manual tasks such as machine operators, cement masons, and dressmakers was not only falling, but falling at an accelerating rate. These jobs fell by 5.6 percent between 1981 and 1991, 6.6 percent between 1991 and 2001, and 11 percent between 2001 and 2011.
²⁶
In contrast, both nonroutine cognitive work and nonroutine manual work grew in all three decades.

C
ONVERSATIONS
WITH
senior executives help explain this pattern in the data. A few years ago, we had a very candid discussion with one CEO, and he explained that he knew for over a decade that advances in information technology had rendered many routine information-processing jobs superfluous. At the same time, when profits and revenues are on the rise, it can be hard to eliminate jobs. When the recession came, business as usual obviously was not sustainable, which made it easier to implement a round of painful streamlining and layoffs. As the recession ended and profits and demand returned, the jobs doing routine work were not restored. Like so many other companies in recent years, his organization found it could use technology to scale up without these workers.

As we saw in chapter 2, this reflects Moravac’s paradox, the insight that the sensory and motor skills we use in our everyday lives require enormous computation and sophistication.
²⁷
Over millions of years, evolution has endowed us with billions of neurons devoted to the subtleties of recognizing a friend’s face, distinguishing different types of sounds, and using fine motor control. In contrast, the abstract reasoning that we associate with ‘higher thought’ like arithmetic or logic is a relatively recent skill, developed over only a few thousand years. It often requires simpler software and less computer power to mimic or even exceed human capabilities on these types of tasks.

Of course, as we’ve seen throughout this book, the set of tasks machines can do is not fixed. It is constantly evolving, just as our use of the word “computer” itself has evolved from referring to a job that humans do to referring to a piece of equipment.

In the early 1950s, machines were taught how to play checkers and could soon beat respectable amateurs.
²⁸
In January 1956, Herbert Simon returned to teaching his class and told his students, “Over Christmas, Al Newell and I invented a thinking machine.” Three years later, they created a computer program modestly called the “General Problem Solver,” which was designed to solve, in principle, any logic problem that could be described by a set of formal rules. It worked well on simple problems like Tic-Tac-Toe or the slightly harder Tower of Hanoi puzzle, although it didn’t scale up to most real-world problems because of the combinatorial explosion of possible options to consider.

Cheered by their early successes and those of other artificial intelligence pioneers like Marvin Minsky, John McCarthy and Claude Shannon, and Simon and Newell were quite optimistic about how rapidly machines would master human skills, predicting in 1958 that a digital computer would be the world chess champion by 1968.
²⁹
In 1965, Simon went so far as to predict, “machines will be capable, within twenty years, of doing any work a man can do.”
³⁰

Simon won the Nobel Prize in Economics in 1978, but he was wrong about chess, not to mention all the other tasks that humans can do. His mistake may have been more about the timing than the ultimate outcome. After Simon made his prediction, computer chess programs improved by about forty points per year on the official Elo chess rating system. On May 11, 1997, forty years after Simon’s prediction, an IBM computer called Deep Blue beat the world chess champion, Gary Kasparov, after a six-game match. Today, no human can beat even a mid-tier computer chess program. In fact, software and hardware have progressed so rapidly that by 2009, chess programs running on ordinary personal computers, and even mobile phones, have achieved grandmaster levels with Elo ratings of 2,898 and have won tournaments against the top human players.
³¹

Labor and Capital

Technology is not only creating winners and losers among those with differing amounts of human capital, it is also changing the way national income is divided between the owners of physical capital and labor (people like factory owners and factory workers)—the two classical inputs to production.

When Terry Gou, the founder of Foxconn, purchased thirty thousand robots to work in the company’s factories in China, he was substituting capital for labor.
³²
Similarly, when an automated voice-response system usurps some of the functions of human call center operators, the production process has more capital and less labor. Entrepreneurs and managers are constantly making these types of decisions, weighing the relative costs of each type of input, as well as the effects on the quality, reliability, and variety of output that can be produced.

Rod Brooks estimates that the Baxter robot we met in chapter 2 works for the equivalent of about four dollars per hour, including all costs.
³³
As we discussed at the start of this chapter, to the extent that a factory owner previously employed a human to do the same task that Baxter could do, the economic incentive would be to substitute capital (Baxter) for labor as long as the human was paid more than four dollars per hour. If output stays the same, and assuming no new hires are made in engineering, management, or sales at the company, it would increase the ratio of capital to labor input.
*

Compensation of the remaining workers could go up or down in the wake of Baxter’s arrival. If their work is a close substitute for the robots’, then there will be downward pressure on human wages. That will grow even worse if Moore’s Law and other advances allow future versions of Baxter to work for two dollars per hour, and then one dollar per hour, and so on, while handling an increasing variety and complexity of tasks. However, economic theory also holds open the possibility that the remaining workers would see an increase in pay. In particular, if their work complements the technology, then demand for their services will increase. In addition, as technical advances increase labor productivity, employers can afford to pay more for each worker. In some cases, this is reflected directly in higher wages and benefits. In other cases, the prices of products and services fall, so the real wage of workers increases as they are able to buy more with each dollar. As productivity improves, total amount of output per person would increase but the amount earned by human workers could either fall or rise, with the remainder going to capital owners.

Of course, almost every economy has been using technology to substitute capital for labor for decades, if not centuries. Automatic threshing machines replaced a full 30 percent of the agricultural labor force in the middle of the nineteenth century, and industrialization continued at a brisk pace throughout the twentieth century. Nineteenth-century economists like Karl Marx and David Ricardo predicted that the mechanization of the economy would worsen the fate of workers, ultimately driving them to a subsistence wage.
³⁴

What has actually happened to the relative share of capital and labor? Historically, despite changes in the technology of production, the share of overall GDP going to labor has been surprisingly stable, at least until recently. As a result, wages and living standards have grown dramatically, roughly in line with the dramatic increases in productivity. In part, this reflects the increases in human capital that have paralleled the more visible increases in equipment and buildings in the economy. Dale Jorgenson and his colleagues have estimated that the overall magnitude of the human capital in the U.S. economy, as measured by its economic value, is as much as ten times the value of the physical capital.
³⁵
As a result, labor compensation has grown along with payments to owners of physical capital via profits, dividends, and capital gains.

Figure 9.3 shows that in the past decade, the relatively consistent division between the shares of income going to labor and physical capital seems to be coming to an end. As noted by Susan Fleck, John Glaser, and Shawn Sprague in the
Monthly Labor Review
: “Labor share averaged 64.3 percent from 1947 to 2000. In the United States, the share of GDP going to labor has declined over the past decade, falling to its lowest point in the third quarter of 2010, 57.8 percent.”
³⁶
What’s more, this is a global phenomenon. Economists Loukas Karabarbounis and Brent Neiman of the University of Chicago find that “the global labor share has significantly declined since the early 1980s, with the decline occurring within the large majority of countries and industries.”
³⁷
They argue that this decline is likely due to the technologies of the information age.

The fall in labor’s share is in part the consequence of two trends we have already noted: fewer people are working, and wages for those who are working are lower than before. As a result, while labor compensation and productivity in the past rose in tandem, in recent years a growing gap has opened.

If productivity is growing and labor as a whole isn’t capturing the value, who is? Owners of physical capital, to a large extent. While the economy remained mired in a slump, profits reached historic highs last year, both in absolute terms ($1.6 trillion) and as a share of GDP (26.2 percent in 2010, up from the 1960–2007 average of 20.5 percent).
³⁸
Meanwhile, real spending on capital equipment and software has soared by 26 percent while payrolls have remained essentially flat, as noted by Kathleen Madigan.
³⁹

What’s more, the collapse in the share of GDP going to labor actually understates how the situation has deteriorated for the typical worker. The official measure of labor compensation includes soaring wages for a small number of superstars in media, finance, sports, and corporate positions. Furthermore, it is debatable that all of the compensation going to CEOs and other top executives is solely due to their ‘labor’ income. It may also reflect their bargaining power, as suggested by Harvard Law Professor Lucian Bebchuk and others.
⁴⁰
In this sense, it might make sense to think of CEOs’ income as due to their control of capital, not labor, at least in part.

While the share of national income to capital has been growing at the expense of labor, economic theory does not necessarily predict that this will continue, even if robots and other machines take over more and more work. The threat to capital’s share comes not (just) from the bargaining power of various types of human labor, from CEOs or labor unions but, ironically, from other capital. In a free market, the biggest premiums go to the scarcest inputs needed for production. In a world where capital can be replicated at a relatively low cost (think of computer chips or even software), the marginal value of capital will tend to fall, even if more capital is used overall. The value of existing capital will actually be driven down as new capital is added cheaply at the margin. Thus, the rewards earned by capitalists may not automatically grow relative to labor. Instead the shares will depend on the exact details of the production, distribution, and governance systems.